xlog-relcomp-probe: XLOG recursion on DR64, and XLOG × egglog over one relation by AdaWorldAPI · Pull Request #1896 · AdaWorldAPI/lance-graph

AdaWorldAPI · 2026-10-31T18:11:55Z

This is stage 1 of D-REASONING-CODEGRAPH-0: XLOG and egglog reasoning over the same stored relation. It lives in the existing xlog-relcomp-probe, as #1888's handoff asks. There is no new crate, engine or store.

XLOG recursion on the CPU kernel (src/recursive.rs)

admit_closure accepts XLOG's compiled recursive SCC for linear transitive closure, and nothing else. That shape is a base Scan edge plus a ChainJoin of path and edge on path.1 = edge.0, with an agreeing fallback.
Non-linear and mutual recursion are refused, each with its own reason.
closure binds edge under the sealed frame, types edge ∘ edge with the existing chain typing, and evaluates semi-naively with the existing relcomp::compose.
support(x, y) gives the facts some derivation of path(x, y) can use, and affected(u, v) the answers a fact can feed. Both are masks computed from the closure on demand; derivations are not stored.

tests/recursive.rs, 9 tests:

the closure matches BFS over the tuples and a naive fixpoint walk of XLOG's own compiled bodies;
a 200-case differential against BFS passes;
cycles terminate;
semi-naive work is exactly one row OR per derived pair;
support covers both diamond branches and excludes off-route facts;
the refusals hold, and a stale frame is refused.

egglog on the same register (tests/shared_relation.rs, 6 tests)

egglog derives equality from (rule ((edge a b)) ((union a b))), with proofs from prove.

egglog's a = b equals XLOG's closure of E ∪ Eᵀ on every pair of a two-component graph.
Every directed path is an egglog equality; the converse fails.
Every fact in an egglog proof is a bit of the shared register. For a directed answer, every such fact also lies in XLOG's support mask.
When a fact is removed and the frame advances, XLOG's bind refuses the old frame. The removed fact is found in the old egglog proof by its address, without re-running egglog.

The one conversion is register bits to egglog fact text: 114 B for 6 edges. egglog's tables and union-find are untouched.

Disable runs. Five guards were disabled one at a time, each turning its test red: the support filter, the reflexive closure in affected, the non-linear guard, the mutual-recursion guard, and semi-naive deltas. Three of these tests first stayed green and were tightened until they discriminated.

Not done:

timing;
the MLIR/DialEgg, SQL and Java legs;
egglog proofs of a directed relation;
recursion shapes beyond one linear closure.

cargo test --release passes and clippy -D warnings is clean.

Generated by Claude Code

…i-naively on DR64 admit_closure accepts exactly one linear recursive SCC (base Scan edge, step ChainJoin path.1 = edge.0 projecting 0 and 3, fallback agreeing) and refuses non-linear and mutual recursion. closure binds edge under the sealed frame and runs semi-naive rounds over the existing relcomp::compose. support and affected read derivations back from the closure as masks on demand. Tests: BFS and a fixpoint walk of XLOG's own plan as references, a 200-case differential, cycle termination, both diamond alternatives in support, the affected set of one fact, refusals, and a stale frame. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

One edge register under a sealed frame is read by XLOG's compiled closure (directed reachability) and by egglog (equality with proofs, via (rule ((edge a b)) ((union a b)))). Tests: egglog equality equals the closure of E u E^T on all pairs; every directed path is an equality; an egglog proof rests only on bits of the shared register, and on a directed answer every proof fact lies in XLOG's support mask; a removed fact invalidates the XLOG bind and is found in the old proof by address. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

…ation stage Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

coderabbitai · 2026-10-31T18:12:04Z

Important

Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

Run configuration

Source mode: Auto
Repository: Unavailable
Branch: Unavailable
File: No configuration file from this repository applied
Commit: Unavailable
Layers: Organization UI
Review profile: CHILL
Plan: Advanced
Run ID: 7f2a7d54-ba8d-47f3-8c1d-b9f861b1b657

Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

Share

trace(edge, valid, x, y) replays closure's semi-naive rounds and returns the derivation evaluation performed: from the round path(x, y) first appeared, back through the previous round's delta and edge, smallest intermediate first. tests/trace.rs checks it against BFS shortest paths, the register, and support. tests/provenance.rs: provenance is source root + frame version + program hash + CE64 declared reading (XLOG only). The root is an observer and its sorted, deduplicated facts, so re-serializing a report is never new evidence; the program hash covers the program, never the facts. Derivation multiplicity is counted separately from independent roots, and two consumers of one root are one observation. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

…, not lineage derivation_multiplicity_is_not_independent_evidence counted the union of roots over all derivations (lineage) and called it independent evidence. With A reporting 1->2->4->5 and B only 1->3->4, the derivations rest on {A} and {A, B}; absorption leaves {A}, so there is one independent support, not two. The test now computes the ATMS label (PosBool provenance over source roots) and pins three cases: one observer, B dependent on A, and B independent of A. Lineage stays as a separate count and cannot tell the last two apart. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

…ment label Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

chatgpt-codex-connector · 2026-10-31T19:01:09Z

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-31T19:02:39.828444Z	`2678550`	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2678550a73

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-10-31T19:02:37Z

+    for (p, w) in path.iter_mut().zip(e.rows) {
+        *p = w & valid;


Mask unassigned source rows when initializing the closure

When the register contains stale bits in an unassigned row, valid only masks destination columns, so those bits become answers even though their source slot is outside the domain. For example, with valid = 0b11 and only edge[63] = 1, closure returns rows[63] = 1; trace(edge, valid, 63, 0) likewise returns a base-fact witness for that unassigned source. Because the relation uses the same axis on both sides, zero rows whose index is not in valid, and apply the same initialization in trace, so the domain mask excludes invalid endpoints consistently.

Useful? React with 👍 / 👎.

claude added 5 commits October 31, 2026 18:07

xlog-relcomp-probe: tighten the support and non-linear refusal tests

9a7b418

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

xlog-relcomp-probe: pin semi-naive work to one row OR per derived pair

38d21a5

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

xlog-relcomp-probe: README section and board entry for the shared-rel…

b75a26c

…ation stage Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

claude added 6 commits October 31, 2026 18:27

xlog-relcomp-probe: record the traceroute and provenance disable runs

8ecc963

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

xlog-relcomp-probe README: independent support is the minimal-environ…

da0feb5

…ment label Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

xlog-relcomp-probe: scope the test-local label as a cold oracle

c295d25

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

board: co-research map for the D-STAUNEN arc (#1885..#1889)

2678550

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01XAqKMNZqKXSaeb9g2ypywF

AdaWorldAPI marked this pull request as ready for review October 31, 2026 19:01

AdaWorldAPI merged commit 3927001 into main Oct 31, 2026
11 checks passed

chatgpt-codex-connector Bot reviewed Oct 31, 2026

View reviewed changes

AdaWorldAPI mentioned this pull request Oct 31, 2026

D-DT64-7: behaviour addressing and registration (Frozen, Learned, Explore); fruits from the D-STAUNEN co-research #1898

Merged

Navigation Menu

Repository navigation

xlog-relcomp-probe: XLOG recursion on DR64, and XLOG × egglog over one relation - #1896

xlog-relcomp-probe: XLOG recursion on DR64, and XLOG × egglog over one relation#1896
AdaWorldAPI merged 11 commits into
mainfrom
ccr-85acaff0-oszq40-reason

AdaWorldAPI commented Oct 31, 2026

Uh oh!

coderabbitai Bot commented Oct 31, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector Bot commented Oct 31, 2026 •

edited

Loading

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Oct 31, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

AdaWorldAPI commented Oct 31, 2026

Uh oh!

coderabbitai Bot commented Oct 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot commented Oct 31, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Oct 31, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

coderabbitai Bot commented Oct 31, 2026 •

edited

Loading

chatgpt-codex-connector Bot commented Oct 31, 2026 •

edited

Loading